Struct GpuArchSpec 

pub struct GpuArchSpec {
    pub name: String,
    pub max_threads_per_sm: u32,
    pub max_blocks_per_sm: u32,
    pub max_warps_per_sm: u32,
    pub warp_size: u32,
    pub registers_per_sm: u32,
    pub register_alloc_granularity: u32,
    pub shared_memory_per_sm: u32,
    pub shared_memory_alloc_granularity: u32,
    pub sm_count: u32,
}

GPU architecture specification.

Fields

name: String

Compute capability name (e.g., “sm_90”, “gfx1100”).

max_threads_per_sm: u32

Max threads per SM/CU.

max_blocks_per_sm: u32

Max blocks (thread groups) per SM.

max_warps_per_sm: u32

Max warps per SM.

warp_size: u32

Warp size (32 for NVIDIA, 64 for AMD RDNA, 32 for AMD CDNA).

registers_per_sm: u32

Total registers per SM.

register_alloc_granularity: u32

Register allocation granularity (registers are allocated in chunks).

shared_memory_per_sm: u32

Shared memory per SM (bytes).

shared_memory_alloc_granularity: u32

Shared memory allocation granularity (bytes).

sm_count: u32

Number of SMs/CUs on the device.

Implementations

impl GpuArchSpec

pub fn hopper() -> Self

NVIDIA Hopper (SM 9.0) — H100.

pub fn ada_lovelace() -> Self

NVIDIA Ada Lovelace (SM 8.9) — RTX 4090.

pub fn ampere() -> Self

NVIDIA Ampere (SM 8.0) — A100.

pub fn cdna3() -> Self

AMD CDNA3 — MI300X.

pub fn generic() -> Self

Generic spec for WebGPU/software simulation.

Trait Implementations

impl Clone for GpuArchSpec

fn clone(&self) -> GpuArchSpec

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for GpuArchSpec

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.